Azzedine Boukerche

Dynamic load balancing strategies for conservative parallel simulations

This paper studies the problem of load balancing for conservative parallel simulations for execution on a multicomputer. The synchronization protocol makes use of Chandy-Misra null-messages. We propose a dynamic load balancing algorithm which assumes no compile time knowledge about the workload parameters. It is based upon a process migration mechanism, and the notion of CPU-queue length, which indicates the workload at each processor.We examine two variations for the algorithm which we refer to as centralized and multi-level hierarchical methods, in the context of queueing network simulation of a torus. The torus was chosen because it of its many cycles aid in the formation of deadlock making it a stress test for any conservative synchronization protocols. Our experiments indicate that our dynamic load balancing schemes significantly reduce the run time of an optimized version of Chandy-Misra null message approach, and decreases by 30-40\% the synchronization overhead when compared to the use of a static partitioning algorithm. Significantly, the results obtained also indicate that the multi-level scheme always outperforms both the centralized load balancing approach and the static partitioning algorithm.

A static partitioning and mapping algorithm for conservative parallel simulations

In this paper, we consider the problem of partitioning a conservative parallel simulation for execution on a multi-computer. The synchronization protocol makes use of null messages [6]. We propose the use of a simulated annealing algorithm with an adaptive search schedule to find good (sub-optimal) partitions. The paper discusses the algorithm, its implementation and reports on the performance results of simulations of a partitioned FCFS queueing network model executed on iPSC/860 hypercube. The results obtained are compared with a random partitioning. They show that a partitioning which makes use of our simulated annealing algorithm results in a reduction of 25-35% of the running time of the simulations when compared to the running time of a random partition of the model.

Analysis of a Randomized Congestion Control Scheme with DSDV Routing in ad Hoc Wireless Networks

Ad hoc wireless networks are expected to play an increasingly important role in future civilian and military settings where wireless access to a wired backbone is either ineffective or impossible. However, frequent topology changes caused by node mobility make routing in ad hoc wireless networks a challenging problem. In addition, limited capabilities of mobiles require a control on node congestion due to message forwarding. In this paper, we present a randomized version, called R-DSDV, of the known destination-sequenced distance vector (DSDV) routing protocol for ad hoc networks and analyze its capabilities for distributed congestion control. The analysis is based on a novel approach that uses Markov chains for representing the distribution of routing information over the network. Our results indicate that a probabilistic congestion control scheme based on local tuning of protocol parameters is feasible and that such a mechanism can be effective in reducing the amount of traffic routed through a node which is temporarily congested. The traffic reduction is almost linear with respect to the decrease in the R-DSDV routing table advertisement rate of the congested node. On an average, the convergence time is relatively small as compared to the transmission rates of protocol messages.

Dynamic Grid-Based Approach to Data Distribution Management

Data distribution management (DDM) is one of the services defined by the DoD High Level Architecture and is necessary to provide efficient, scalable mechanisms for distributing state updates and interaction information in large scale distributed simulations. In this paper, we focus on data distribution management mechanisms (also known as filtering) used for real time training simulations. We propose a new method of DDM, which we refer to as the dynamic grid-based approach. Our scheme is based on a combination of a fixed grid-based method, known for its scalability, and a region-based strategy, which provides greater accuracy than the fixed grid-based method. We describe our DDM algorithm, its implementation, and report on the performance results that we have obtained using the RTI-Kit framework. Our results clearly indicate that our scheme is scalable and that it reduces the message overhead by 40%, and the number of multicast groups used by 98% when compared to the fixed grid-based allocation scheme using 10 nodes, 1000 objects, and 20,000 grid cells.

SWiMNet: a scalable parallel simulation testbed for wireless and mobile networks

We present a framework, called SWiMNet, for parallel simulation of wireless and mobile PCS networks, which allows realistic and detailed modeling of mobility, call traffic, and PCS network deployment. SWiMNet is based upon event precomputation and a combination of optimistic and conservative synchronization mechanisms. Event precomputation is the result of model independence within the global PCS network. Low percentage of blocked calls typical for PCS networks is exploited in the channel allocation simulation of precomputed events by means of an optimistic approach.Various experiments were conducted to study the performance and scalability of SWiMNet using a realistic mobility model and executed on a cluster of workstations. Experimental results indicate that our parallel simulation model yields good speedup, and significantly reduces the execution time compared to a sequential implementation. Finally, an analytical study of our PCS simulation model is also presented and compared with the experimental results. Our model is found to be consistent with the analytical study.

Exploiting model independence for parallel PCS network simulation

We present a parallel simulator (SWiMNet) for PCS networks using a combination of optimistic and conservative paradigms. The proposed methodology exploits event precomputation permitted by model independence within the PCS components. The low percentage of blocked calls is exploited in the channel allocation simulation of precomputed events by means of an optimistic approach. Experiments were conducted with various call arrival rates and mobile host densities on a cluster of Pentium workstations. Performance results indicate that the SWiMNet achieves a speedup of 6 employing 8 workstations, and a speedup of 12 with 16 workstations.

Dynamic grid-based multicast group assignment in data distribution management

Data distribution management (DDM) is one of the services defined by the DoD High Level Architecture. DDM is necessary to provide efficient, scalable mechanisms for distributing state updates and interaction information in large scale distributed simulations. We describe data distribution management mechanisms (also known as filtering) used for real time training simulations. We propose a new DDM approach to multicast group allocation, which we refer to as a dynamic grid-based allocation. Our scheme is based on a combination of a fixed grid-based method, known for its low overhead and ease of implementation, and a sender-based strategy, which uses fewer multicast groups than the fixed grid-based method. We describe our DDM algorithm, its implementation, and report on the performance results that we have obtained using the RTI-Kit framework. These results include the outcome of experiments comparing our approach to the fixed grid-based method, and they show that our scheme is scalable and significantly reduces the message overhead of previous grid-based allocation schemes.

A distributed graph algorithm for the detection of local cycles and knots

In this paper, a distributed cycle/knot detection algorithm for general graphs is presented. The algorithm distinguishes between cycles and knots and is the first algorithm to our knowledge which does so. It is especially relevant to an application such as parallel simulation in which 1) cycles and knots can arise frequently 2) the size of the graph is very large, and 3) it is necessary to know if a given node is in a cycle or a knot. It requires less communication than previous algorithms-2m vs. (at least) (4m) for the Chandy and Misra algorithm, where m is the number of links in the graph. It requires O (nlog (n)) bits of memory, where n is the number of nodes. The algorithm differs from the classical diffusing computation methods through its use of incomplete search messages to speed up the computation. We introduce a marking scheme in order to identify strongly connected subcomponents of the graph which cannot reach the initiator of the algorithm. This allows us to distinguish between the case in which the initiator is in a cycle (only) or is in a knot.

A distributed algorithm for dynamic channel allocation

Recent demand for mobile telephone service has been growing rapidly while the electro-magnetic spectrum of frequencies allocated for this purpose remains limited. Any solution to the channel assignment problem is subject to this limitation, as well as the interference constraint between adjacent channels in the spectrum. Channel allocation schemes provide a flexible and efficient access to bandwidth in wireless and mobile communication systems. In this paper, we present an efficient distributed algorithm for dynamic channel allocation based upon mutual exclusion model, where the channels are grouped by the number of cells in a cluster and each group of channels cannot be shared concurrently within the cluster. We discuss the algorithm and prove its correctness. We also show that the algorithm requires at most (worst case) O(Ng⋅Nn log Nn) messages, where Ng is the number of groups and Nn is the number of neighbors. This is compared to Choys algorithm which requires O(Ng2⋅Nn), where Ng is the number of groups and Nn is the number of neighboring cells in the system. We report our algorithms performance with several channel systems using different types of call arrival patterns. Our results indicate that significant low denial rate, low message complexity and low acquisition time can be obtained using our algorithm.

Performance comparison of data distribution management strategies

Abstract: Data Distribution Management (DDM) is a High Level Architecture/Run-time Infrastructure (HLA/RTI) service that manages the distribution of state updates and interaction information in large scale distributed simulations, limits and controls the volume of data exchanged during the simulation. In this paper we focus upon the following three DDM schemes: Region-Based, Fixed and Dynamic Grid-Based techniques. In an effort to determine the most efficient model for applying the DDM service, we developed a mini-RTI Kit framework and propose a set of benchmarks to compare these DDM schemes. Using the RTI Kit, we performed extensive simulation experiments and present our analysis of the results of numerous federation executions.